AutoScout24 speeds up development 10x with OpenAI's Codex and ChatGPT

OpenAI has published a case study on AutoScout24, in which the car marketplace describes how it built ChatGPT and Codex into its daily work. The company says that for some projects this cut the development cycle from 2–3 weeks to 2–3 days and helped scale AI far beyond a single engineering team.

Why AutoScout24 needed AI

AutoScout24 Group runs a large car-sales platform in Europe and Canada: more than 30 million users a month, over 2 million listings and a network of 45,000 dealer partners. At that scale, any delay in development quickly becomes a business problem. The team has to support several brands at once, develop the product for buyers and dealers, and keep stable systems that have already accumulated complex internal logic.

The company says its earlier, piecemeal process improvements were no longer enough. Engineering teams were under pressure from migrations, legacy systems and a growing demand for new features, while the cost of slow releases kept rising. So AutoScout24 decided to use large language models not as a standalone experiment but as a way to rebuild its approach to development, testing and internal collaboration between product, data and engineering. The goal was no longer to speed up individual tasks but to remove systemic friction from the entire product delivery chain.

How the tools were rolled out

The rollout had two layers. First, ChatGPT was deployed to almost the entire organization, giving about 2,000 employees access and creating a baseline of AI literacy outside engineering. In parallel, Codex was built into the daily processes of those who build the product directly: engineering, product and data teams. In total, this covers roughly 1,000 employees in product-building roles, for whom AI is not a separate browser tab but part of everyday operations.

Before that, the company spent three months comparing the tool across different teams and ultimately chose Codex for its ease of use, compatibility with existing processes, and measurable gains in productivity and code quality.

So that adoption would not depend solely on top-down orders, AutoScout24 built a network of AI Champions: internal guides from different functions. They collect feedback, help translate model capabilities into clear scenarios, and make sure AI is embedded in existing processes rather than living as a separate pilot.

"Codex has become one of the key tools in our engineering processes, delivering measurable gains in productivity, quality and speed," said

Frederik Kraus, CTO of AutoScout24 Group.

Where the results showed up

The company saw the fastest effect where there used to be a lot of manual routine, repetitive operations and long approval cycles. AI is used not for abstract experiments but for specific tasks that slow down shipping changes every day. As a result, the tools help speed up iterations, take load off experienced developers and improve consistency across teams without a total overhaul of architecture or processes. These are the areas where the effect is easiest to measure in days, reviewer workload and decision speed.

• Automated pull request reviews
• Large refactoring tasks
• Preparing technical documentation
• Post-outage incident analysis
• Prototyping ideas outside engineering teams

According to AutoScout24, timelines on some projects fell roughly tenfold, from 2–3 weeks to 2–3 days. Engineering throughput also rose, and code review and documentation now depend less on manual work. The company notes a separate effect among non-technical employees, who can now test hypotheses and build simple prototypes themselves.

The next step is to embed AI more deeply into core internal systems and into customer-facing products for buyers and dealers.

What this means

This case shows that enterprise AI is increasingly judged not by the number of licenses purchased but by how much it shortens the path from idea to release. If a company can combine broad access, embedded workflows and clear owners of the rollout, ChatGPT and Codex turn from an experiment into a real layer of production infrastructure. That is what separates a working deployment from a fashionable pilot. And here, numbers matter more than slogans.